Multispectral ghost imaging is a perfect combination of spectral and ghost imaging. Compared to single-band ghost imaging, multispectral ghost imaging provides richer information, which may improve the accuracy and reliability of object recognition, material analysis, and medical diagnosis. Traditional multispectral ghost imaging is accomplished by employing multiple detectors or time-divided detection via filter wheeling to capture images of various spectral channels and combine them into a color image. These strategies have limitations, such as complex structures, and require a large amount of data. Alternatively, by encoding and decoding in the projection and recovery stages, respectively, multispectral multiplexed ghost imaging can obtain the multispectral information of objects in one projection cycle. However, there are still some problems in existing multispectral multiplexed ghost imaging, such as requiring a compressed sensing algorithm in the restoration phase, significantly increasing time consumption. In this study, we propose cosinusoidal encoding multiplexed multispectral ghost imaging, which can effectively improve the imaging efficiency of multispectral ghost imaging while ensuring high-quality imaging through a simple and flexible encoding strategy.
The process of cosinusoidal encoding multiplexed multispectral ghost imaging is divided into three steps. First, Hadamard basis patterns and cosinusoidal encoding matrices are combined to generate colored illumination patterns. The second step is similar to traditional ghost imaging; the constructed colored patterns are used to illuminate the target object, and a single-pixel detector is used to collect the reflection signal of the object. Subsequently, the conventional linear algorithm was used to obtain a composite grayscale image. The third step involved reconstructing the multispectral image. During this phase, a composite grayscale image of the object is converted to Fourier space for recombining the information of each channel. Further, the spectral information of the different channels is acquired separately and combined to generate a multispectral image. The influence of the ideal, Gaussian, and Butterworth low-pass filters on the restoration quality of multispectral images was simulated, and the approach was compared with the traditional method.
The effectiveness of the proposed approach was verified by the numerical simulations of colored objects. In the restoration phase, different filtering parameters were used to extract the reconstructed spectral component information, and the PSNR (peak signal to noise ratio) and SSIM (structural similarity) were used to evaluate the reconstructed image quality (Fig. 2). The simulation results of the resolution test image show that multispectral images reconstructed with an ideal low-pass filter exhibit a ringing phenomenon. In addition, the Gaussian low-pass filter and Butterworth low-pass filter can improve image quality. When the filter radius is small, the reconstructed image becomes blurred. With an increased filter radius, the quality of the reconstructed images improves and gradually reaches a maximum value. As the filter radius increases, the image restoration quality rapidly decreases owing to the influence of other spectral component information (Fig. 3). Simulations indicate that for the images of natural scenery with higher gray levels, the Gaussian low-pass filter has the best restoration performance for a small filter radius, followed by the Butterworth low-order filter, the Butterworth high-order filter, and then ideal filters. The inflection point of the image quality-change curve corresponding to the filter with a better restoration effect appears earlier (Fig. 4, Fig. 5). The simulation results show that the Gaussian low-pass filter has certain advantages over the ideal and Butterworth filters. Preliminary experimental results demonstrate that the proposed method can achieve better imaging results when a Gaussian filter with a smaller filter radius is used. Through comparative experiments with traditional methods, the effectiveness of the proposed method and rationality of the filtering strategy are verified.
This paper demonstrates an approach for combining cosinusoidal encoding and multispectral ghost imaging. It can flexibly construct cosine coding matrices to meet specific needs and then combine them with the Hadamard basis patterns to project the imaging object with color illumination light. Each channel is encoded independently by the cosine coding matrices, and the information of the component images can be decoded in the frequency domain using the Fourier frequency shift in the cosine structure coding matrices. This approach requires only a single-pixel detector to obtain a multispectral image of the target object, simplifying the multispectral imaging system. Simultaneously, using the compressed sensing optimization algorithm is not required in the restoration process, which reduces the computational time of image reconstruction and improves the imaging efficiency. Numerical simulations and experimental results show that the new approach can effectively improve the efficiency of multispectral information acquisition and shorten the image reconstruction time. In frequency-domain filtering, the Gaussian low-pass filter is more advantageous than the ideal and Butterworth filters.
